C OMPOSITIO M ATHEMATICA
G IUSEPPE V IGNA S URIA
q-pseudoconvex and q-complete domains
Compositio Mathematica, tome 53, n
o1 (1984), p. 105-111
<http://www.numdam.org/item?id=CM_1984__53_1_105_0>
© Foundation Compositio Mathematica, 1984, tous droits réservés.
L’accès aux archives de la revue « Compositio Mathematica » (http:
//http://www.compositio.nl/) implique l’accord avec les conditions gé- nérales d’utilisation (http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une in- fraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
105
q-PSEUDOCONVEX
andq-COMPLETE
DOMAINSGiuseppe Vigna
Suria© 1984 Martinus Nijhoff Publishers, Dordrecht. Printed in The Netherlands.
Introduction
The Levi
problem
wasoriginally posed
in thefollowing
terms: if D is adomain in C " with
e2 boundary
which ispseudoconvex
is D a domain ofholomorphy?
It was then realised that the
hypothesis
on theboundary
can beremoved if
pseudoconvexity
isreplaced by completeness,
which is aconcept that makes sense in any
analytic manifold,
and the final solution of the Leviproblem
due to Grauert says that acomplete analytic
manifold is
necessarily
Stein[3].
The
original spirit
of theproblem
has not beenbetrayed:
domainswith
C2 boundary
in C" arepseudoconvex
if andonly
ifthey
arecomplete ([4]
p.50).
The same is not true any more if Cn is
replaced by
anyanalytic
manifold: a well known
example
of Grauertprovides
a subset withC2 boundary
of acomplex
torus which ispseudoconvex
but allholomorphic
functions thereon are constant.
In this paper we prove that a
q-pseudoconvex
open subset of a Stein manifold isnecessarily q-complete (the
converse is also true, see[2]).
Thisseems to be one of those facts that every
complex analyst believes, perhaps
forpsycological
reasons, but noprecise
referenceis,
to myknowledge,
available and all mathematicians whom 1 have asked so far don’t seem to know how aprecise proof
should go; the modest aim of this paper is to fill this gap andprovide
a definite reference.Most of the ideas in the
proof
are due to Mike Eastwood to whom 1am, once more,
deeply grateful.
We
briefly
recall the basic definitions:DEFINITION 1: Let D be an open subset of an
analytic
manifold M ofdimension n ; we say that D has
C2 boundary
if for all x E aD there existsan open
neighbourhood
U of x and aC2
function (p:U ~ R,
calleddefining function
of D at x s.t. D ~ U =(y
E U s.t.cp(y) 01 -
and106
d~(x) ~ 0;
in these conditions we can consider thecomplex
Hessianwhere zl, z2, ... , zn are local
holomorphic
coordinates at x. Thesignature
of this Hermitian matrix does not
depend
on the choice of the localholomorphic
coordinates but it doesdepend
on cp. However the Leviform
where
TxaD
={v
=03A3m1v1~/~zl ~ Tx M
s. t.03A3n1vi~~/~zi(x)
=0}
is the holo-morphic
tangent space of aD at x, has asignature
thatdepends only
onD and x.
If
n ( x )
denotes the number ofnegative eigenvalues of F(~)(x)
we say that D isq-pseudoconvex
ifn(x)
q for all x E aD.DEFINITION 2: A
complex
n-dimensional manifold D is said to beq-complete
if we can find aq-plurisubharmonic
exaustionfunction
on D i.e.a
C2
function 03A8 : D ~ R s.t.(1)
for all c ~ R the setBc = {x ~ D
s.t.03A8(x) c}
isrelatively compact in
D and(2)
Thecomplex
HessianH(03A8)(x)
has atleast n - q positive eigen-
values for all x in D.
0-pseudoconvex
and0-complete
domains aresimply
calledpseudocon-
vex and
complete.
THEOREM:
If D is a
domain withC2 boundary
in a Steinmanifold
M and Dis
q-pseudoconvex
then it is alsoq-complete.
PROOF: We shall divide the
proof
into several steps.Step
1: As there isalways
ananalytic embedding
of M into CN,
for somelarge
N(see [5]
p.359)
we can suppose at once that M is ananalytic
submanifold
of CN.
Choose aholomorphic
tubularneighbourhood
p : V~ M and set
D = p-1(D) (cfr. [1] proof
of Lemma1,
p.131).
We claimthat,
aftershrinking
if necessary,(a)
~x ~ah n V, aD
isC2
at x,(b)
If we considerD
as an open subsetof CN
thenn ( x, ) =
n(p(x), D),
for al x E~
n V.Indeed,
since theproblem
is local we can suppose that local coordi- nates z1, z2, ...,zN have been chosen s.t., near x, M =( z
S.t. zN-n+1 = ZN-n+2 =... = zN= 0),
Zl’ Z2"" Zn are local coordinates of M at x andp Z1, z2, ... , ZN ) = (z1,
z2, ... , zn,0,...,0).
Let
Û
be aneighbourhood
of xin CN
so small that zl, z2,,..., zN are defined inÛ
and that there exists aC2 defining
function (D : U =~
M - R for D withd03A6(x) ~
0 andU c
V.By shrinking Û
if necessary wecan also suppose that
~ p-1(U).
Define : ~ R by = 03A6 o p
i.e.(z1,
z2, ... ,zN) = 03A6(z1,
z2, ... , 1 z,,,0, ... , ,0). Then
is adefining
function forb
at x.Moreover
where as usual v =
03A3Ni=1vi~/~zi,
andThis proves the claim.
Step
2:So,
if we suppose that D isq-pseudoconvex
we havethat,
Vx OE V ~
aD, ab
ise2
at x andn(x, )
q.Consider the function p :
CN ~ R given by
Where dist denotes the Euclidean distance. Since Vx E=-
aD, aD
isC2
at x, we can conclude that there exists aneighbourhood Û’
of aDin CN
onwhich p is
C2 (by
the inverse functiontheorem).
By shrinking Û’
if necessary, we can also suppose thatVy
inU’
there existsexactly
onepoint c(y) ~ aD n U’
which is the closestpoint
to yunder the Euclidean
distance,
thatd03C1(c(y))
~ 0 and thatn(c(y), )
q.Let (P:
Û’ ~ ~ R
be the function (p =log
p; we claim that the Hessian(£cp)(y)
has at most qpositive eigenvalues B;/y.
Indeed suppose that this is
false,
i.e. there exists apoint y
inU’
nD
s.t.
(H~)(y)
has(at least) q + 1 positive eigenvalues;
thegeometric interpretation
is: there are linear coordinates(t1, t2l 1 N) of CN
s.t. theHermitian form
given by
the matrix108
is
positive
definite on the linearsubspace V
ofTvCN = CN spanned by
(~/~1, ~/~t2,...,~/~tq+1).
By Taylor’s
theorem we havewhere a, = 1 2~~/~ti(y)
andbjk
=~2~(y)/~tj~tk
are constants, and0(|t|2)
has the
property
thatlim t ~00(|t|2)/|t|2
= 0 and so alsoIn order to
simplify
notation omit the limits of the summands and writeA(t)
= y +03A3tj~/~tj , B(t)
=exp(03A3aiti
+Ebjktjtk)’
Then the above
equality
can be written as03C1(A(t)) - 03C1(y)(B(t)| =
{exp03A3Cjktjtk
+0(|t|2)) -1} 03C1(y)|B(t)| = {03A3Cjktjtk
+0’(|t|2)}03C1(y)|B(t)|,
where the last
equality
is obtainedby expanding
inTaylor
series thefunction exp and
0’(|t|2)
has the sameproperties
as0(|t|2).
Then one hasso we can choose E > 0 small
enough
s. t.~t, |t|
~, one has(a) A(t) ~ ~ ’
and(b) p (A (t» - 03C1(y)|B(t) > 03C1 (y)/2 · 03A3Cjktjtk.
Set u =
c(y)-y
and define ananalytic
function T on the open ballB~ = {t ~ Cq+1 S.t. |t| ~}:
We can also suppose that E is so small that
T( t )
EU’
if t ~B~.
Then itis easy to
check,
and apicture
showshow,
that if te B,
one has(c) 03C1(T’(t)) 03C1(A(t)) - |u||B(t)| |u|/203A3Cjktjtk 0
This in
particular
proves thatT( t ) E D
for all t EB~ - {0}, and,
since03C1(T(0)) = 03C1(c(y0) = 0,
0 is a minimum for the function poT : B~ ~ R,
and so,
taking partial derivatives,
Using
the chain rule and the fact that T isanalytic
we have:(d) 03A3Nh=1~03C1/~zh(c(y))~Th/~tj(0) = 0 for j = 1,2, .., q + 1.
In other words the vectors
~T/~tj (0), j
=1, 2,...,
q + 1, are inTc( v)~.
Moreover,
B;/tin Cq+1,
we have(e) 03A3q+1j,k=1~203C1o T(0)/~tj~tktjtk |u|/403A3q+1j,k=1Cjktjtk.
To prove this we first observe that it is
clearly enough
to check it forsmall 1 t 1.
From the above
inequality (c), using Taylor series,
we deducefor all
t ~ B~,
wheredjk
=~203C1o T(0)/~tj~tk
are constants and0"(ltI2)
hasthe same
properties
as0(|t|2).
Then,
afterreducing E
if necessary, wehave,
Vt E=-B,,
Let
/J
=ei03B8tj
for0 03B8 2 03C0; writing t’
in the aboveinequality
andobserving
that the second and third term areunchanged
under thesubstitution t -
t’,
wededuce, V0,
and
by choosing B
so that the first term isnegative
we prove theinequality (e).
Using again
the chain rule and the fact that T isanalytic
we have thatthe Hermitian form
is
positive
definite.It follows
easily
that the Hermitian form(~203C1(c(y))/~zh~zm)Nh,m=1
ispositive
definite on the linearsubspace
V of7c(v)~ spanned by
the110
vectors
~T/~Tj(0), j = 1, 2,..., q + 1 ;
inparticular
it follows automati-cally
that these vectors arelinearly independent,
so thatdimV = q
+1;
but
since - p is a defining
function for D atc(y),
we have thatn(c(y), ) q + 1
and this contradicts ourhypothesis,
so that the claim isproved.
Step
3:By restricting ~
toÛ’
n D we find aC2 function,
calledagain
~:W = ’ ~ D ~ R
s.t.(a) limy~~D~(y) = - ~,
(b) (,)glip)(y)
has at most qpositive eigenvalues Vy
in W.Let F be a closed subset of M s.t. D - W ~ int F c F c
D,
and let0 03A8 1 be a C2 bump
function s.t. ’1’ = 0on F, ’1’
= 1 in aneighbour-
hood of M -
D,
and supppse that F is chosen sothat ~(y)
0for y 5É
F.By considering
the functionqq’
= ~ ’P,
we have that(a) limy ~~D~’(y) = - ~,
(b) (H~’)(y)
has atmost q positive eigenvalues ~y ~
D -F, (c) w’
0.Now we use the fact that M is Stein and so
0-complete (see [5],
lemmap.
358)
i.e. there exists a0-plurisubharmonic
exhaustion function À : M - R .’fin E
Z,
the setKn
={y ~
M s.t.03BB(y) n}
is compact, therefore so isF n
Kn
and there exist constantsCn
s.t.Now choose a
C2
functionf :
R - R with theproperties
First we notice
that, Vc E=- R, Be = {y ~
D s. t.x(y) cl
iscontained, by
the property( c)
ofqq’
inf y
E D s.t.f o 03BB (y) c}
which iscompact by
theassumptions
onf
and À.Moreover Bc
is closed in Dand,
sincelimy~~D’(y)
= - oo, it is also closed in M.Thus Bc
iscompact and X
isan exhaustion function.
For
all y E
D we havewhere A(y) = (~03BB/~zi(y) · ~03BB/~zj(y))ni,j=1 is
asemipositive
Hermitianform. If
y E D - F
then there exists a linearsubspace
of7§,D,
ofdimension n - q
where -(H~~’)(y)
ispositive
semidefinite.Therefore
(H~)(y)
ispositive
definite on V.If y E
F theneither y E Ko
~ F in which caseor y E
(Kn+1 - Kn )
~ F for someinteger n 0,
in which caseTherefore X
is alsoq-plurisubharmonic
and we canfinally
say that thetheorem is
proved.
0References
[1] K. DIEDERICH and J.E. FORNAESS: Pseudoconvex domains: bounded strictly plurisub-
harmonic exhaustion functions. Inv. Math. 39 (1977) 129-141.
[2] M.G. EASTWOOD and G. VIGNA SURIA: Cohomologically complete and pseudoconvex
domains. Comm. Math. Helv. 55 (1980) 413-426.
[3] H. GRAUERT: On Levi’s problem and the embedding of real analytic manifolds. Ann.
Math. 68 (1958) 460-472.
[4] L. HORMANDER: An introduction to complex analysis in several variables. Princeton N.J.
Van Nostrand (1966).
[5] R. NARASIMHAN: The Levi problem for complex spaces. Math. Ann. 142 (1961) 355-365.
(Oblatum 2-XI-1982) Giuseppe Vigna Suria Dip. di Matematica Fac. di Scienze Università di Trento 38050 Povo (TN) Italy